3D Scene Layout Generation
Overview
3D scene layout generation studies how to place and organize objects so a scene matches an input while remaining geometrically and functionally plausible. Methods include diffusion and autoregressive models, scene graphs and domain-specific languages, and language- or vision-language agents that build layouts step by step. Many systems use collision checks, physics simulation, reachability constraints, or visual feedback to refine placements, and are tested on indoor-scene benchmarks, image-based reconstruction, and simulation tasks. A central challenge is balancing semantic fidelity and visual quality with physical feasibility and usability; fine-grained evaluation of these properties remains difficult.
Latest papers
80 papers10 papers in the last four weeks, up 43% on the four weeks before. 0.1% of all new papers.